Within the realm of political communications, there is an ongoing debate on the effect on Partisan Selective Exposure (PSE) on fundamental political phenomena such as polarisation. Traditionally, PSE has been measured using self-reported media consumption data. However, limitations have been demonstrated with the use of self-reported data in research on media consumption and these limitations could potentially be contributing to the inconsistent findings regarding PSE. To address this methodological limitation, this study introduces an innovative computational method for measuring PSE based on participant's consumed digital news content
Using web-tracking and survey data from 829 Spanish respondents during the 2023 Spanish elections, the study analyses 22.000 digital news articles containing political content. Employing High Performance Computing (HPC) and the Llama 3.1 70b Large Language Model, the sentiment and stances toward key political actors in these articles are assessed. PSE is then calculated by evaluating the alignment of the content with each respondent’s ideological positioning along the left-right and liberal-conservative continua, while controlling for engagement metrics such as article length and time spent reading. Finally, the study examines the correspondence between the PSE derived from content analysis of consumed digital news and self-reported digital news consumption. This comparison provides insights into the reliability of self-reported data in digital media and selective exposure research. By overcoming the biases of traditional self-report methods, this computational approach advances the study of PSE offering a more accurate understanding of its antecedents and effects.